R2CTA: Reinforcement Learning and Reservoir Computing based Chiplets TSV Assignment

Chiplets based design is gaining more attention in academia because of its success in industry. Its unique structure poses new challenges in the field of physical design optimization. Although many floorplan and TSV assignment algorithms are proposed for dies and 3DICs design optimization, new studies are needed to address the challenges of chiplet systems. With the recent success of artificial intelligence, algorithms should be developed to optimize the design of chiplet systems. This paper proposes a novel approach to the TSV assignment problem in heterogeneous chiplet systems using reinforcement learning (RL) and reservoir computing (RC), coined R2CTA: Reinforcement Learning based Reservoir Computing Chiplet Assignment. R2CTA aims to optimize key metrics of heterogeneous chiplet systems by fine tuning the TSV assignment.